Semantic scene completion is the task of jointly estimating 3D geometry and\nsemantics of objects and surfaces within a given extent. This is a particularly\nchallenging task on real-world data that is sparse and occluded. We propose a\nscene segmentation network based on local Deep Implicit Functions as a novel\nlearning-based method for scene completion. Unlike previous work on scene\ncompletion, our method produces a continuous scene representation that is not\nbased on voxelization. We encode raw point clouds into a latent space locally\nand at multiple spatial resolutions. A global scene completion function is\nsubsequently assembled from the localized function patches. We show that this\ncontinuous representation is suitable to encode geometric and semantic\nproperties of extensive outdoor scenes without the need for spatial\ndiscretization (thus avoiding the trade-off between level of scene detail and\nthe scene extent that can be covered).\n We train and evaluate our method on semantically annotated LiDAR scans from\nthe Semantic KITTI dataset. Our experiments verify that our method generates a\npowerful representation that can be decoded into a dense 3D description of a\ngiven scene. The performance of our method surpasses the state of the art on\nthe Semantic KITTI Scene Completion Benchmark in terms of geometric completion\nintersection-over-union (IoU).\n